A corpus of Nluu

A small corpus of Nluu

Nluu is an endangered language of the Tuu language family from Afrika. The language is not being actively spoken since its handful of speakers (2 as of 2020) live in different villages and do not have direct contact to each other. Nluu has one of the most complex sounding phonetic inventories in the world with over 100 different phonemes, 45 of which are clicks.

A transcribed recording session of Nluu was kindly provided to us by Alena Witzlack-Makarevich (of Hebrew University of Jerusalem). I have converted this corpus to a CSV (comma-separated values) file that can be easily imported to R.

Let’s load the corpus and inspect it!

corpus <- read_csv("nuu.csv",  col_types = cols())
corpus

inspect the corpus data and describe the meaning of the individual columns as well as the structural relationships between them. What are the objects of description in this dataset? Sketch a simple entity-relationship diagram for this dataset and insert it as an image.

Inspectign individual entries

Extract the 24th utterance and count it’s elements (words and morphemes). Write down this utterance as a glossed example.

Part of speech exploration

what are the 5 most frequent part of speech types in the corpus?

sort(table(corpus$ps),decreasing = TRUE)[1:5]

 pro    n part  vtr vitr 
5955 4668 4354 3135 2505 

find the 10 most frequent nouns

sort(table(n_list$word),decreasing = TRUE)[1:10]

      ki ǀʼhuunsi    gǁain      gao    tyuin     ǀoba    ǃqhaa 
     155      120       86       59       57       50       46 
     ǂoo    nǁang   ǀʼhuun 
      45       43       43 

add a new column to the dataset, stem_lexical_type. It should be “pro” for pronouns, “n” for nouns, “v” for verbs and “other” for everythign else.

corpus <- mutate(corpus,stem_lexical_type =case_when(
  ps == 'pro' ~ "pro",
  ps == "n"  ~ "n",
  ps %in% c("vtr", "vitr","vatr") ~ "v",
  TRUE ~ "others"
))
corpus

TODO (advanced, only if you want a challenge!): and now something more difficult — do the same as above, but at the word level and not the stem level! Call this column word_lexical_type

Noun to verb ratio

Noun to verb ratio is a number that tells us how many nominal elements speakers produce in relation to verbal elements — or roughtly, how many nouns are there per sentence. This is a measure of referential density, or how much specific information speakers tend to give per unit of event or state description. While in some languages/cultures one expects to be rather detailed here, so with a high noun to verb ratio, (e.g. ‘A girl wearing a flowery bonet entered the bookstore with large windows and smiled at the old bookkeeper who was just enjoying his morning coffee’) where in some other language the same information could be conveyed as (‘came in, smiled’).

Noun to verb ratio is computed as follows (the \(+1\) in the denominator to prevent division by zero):

\[ N2V = \frac{n(Noun)}{n(Verb)+1} \]

Nouns here may or may not include other elements such as pronouns or demonstrative pronouns. We compute N2V per utterance.

#compute the N2V for all the utterances in the corpus, once taking both pronouns and nouns, and once takign only nouns into account. The output here should be a table with columns record.id, speaker, N2V.NPro and N2V.N, with one row per utterance.

obtain the N2V summaries (median and standard deviation) of N2V per speaker. The output should be a table with columns speaker, N2V.NPro.median, N2V.NPro.sd, N2V.N.median, N2V.N.sd

#produce a violin plot visualisation comparing the N2V ratios per speaker. Only take the N2V for nouns (ignoring pronouns) and restrict the data to only the most common speakers (cutoff criteria on your discretion)

ggplot(re_den_table, aes(x=speaker, y= N2V.N)) +
  geom_point() +
  geom_violin(aes(fill=factor(speaker)))